Reduced feature set for emotion recognition based on angle and size information

Abstract

The correct interpretation of facial emotions is important for many applications like psychology or human-machine interaction. In this paper, a novel set of features for emotion classification from images is introduced. Based on landmark points extracted from the face, angles between point-connecting lines and size information of mouth and eyes are extracted. Experiments compare the quality and reliability of the feature set to landmark-based features and facial action unit based features.